The measurement of the peak value of a repetitive signal plays an important role in different applications, above all in the field of telecommunications. Said task is entrusted to circuits that assume the name of peak detectors. They are used, for example, in gain-control feedback systems in radio receivers and in optical-fibre connections, or else in power amplifiers with “envelope elimination and restoration”.
Peak detectors can be effectively used also for protection of power transistors from the high overvoltages that arise in the case of high output mismatch. A circuit of this sort must be integrated together with the power transistor, since it has the task of monitoring the effective maximum voltage at the output terminal (drain/collector) of the transistor and producing a control signal for activating the protection circuitry readily and in a precise way.
However, traditional solutions suffer from various problems, such as the ones listed below:                the crest factor of the voltage at the output terminal (drain/collector) can vary considerably as a function of the phase of the load reflection coefficient; however, known circuit topologies are very sensitive to the effective waveform of the voltage to be detected, in effect producing an output that is all the more underestimated, the higher the crest factor of the voltage to be measured; and        the output signal is a non-linear function of the input peak value.        
The above phenomena lead to the need for overprotection of the power transistor, causing a sensible drop in the nominal performance. In addition, the same phenomena are more accentuated in the case of MOSFET (Metal-Oxide Semiconductor Field-Effect Transistor) implementations having the topologies traditionally adopted. In addition to this, the design of said circuits is rendered particularly critical in the case of high-frequency signals and/or in the presence of constraints of power dissipation.
The solutions proposed in the literature for peak detection normally comprise an element for rectifying the input voltage and a capacitor for holding the peak value.
Described in J. Millman, et al., “Microelettronica”, McGrawHill (1994), p. 60, is a first peak-detector circuit, reproduced in FIG. 1a. The rectifier element is in this case represented by a diode D1, the current iO of which depends upon the voltage applied according to the function indicated by fD. The accumulation element is instead represented by a capacitance C. The generator IDIS has the purpose of discharging the capacitance C and determines how rapidly the circuit detects negative variations of the peak value of the signal. In said topology, the charge current iO of the capacitance C comes from the input signal, which can constitute a problem because it makes necessary oversizing of the upstream signal-conditioning stages.
In order to solve this problem, described in R. G. Meyer, “Low-Power Monolithic RF Peak Detector Analysis”, IEEE J. Solid-State Circuits, vol. 30, pp. 65-67, (1995), is a second type of peak detector, reproduced in FIG. 1b, in which the diode D1 is replaced by a generic transistor T1. Operation is similar to that of the first circuit, but in this second case the charge current iO no longer comes from the input, thanks to the current gain of the transistor T1.
Both of the circuits are used with a weak discharge current; consequently, it is justifiable to neglect the residual ripple of the output voltage and to make vO(t) to coincide with its mean value VO. On said hypothesis, it is possible to obtain an implicit and general relation between the input waveform and the output voltage. In fact, in periodic conditions, the temporal average of the charge current must coincide with the discharge current:
                              I          DIS                =                                            1                              2                ⁢                π                                      ⁢                                          ∫                                  α                  ⁢                                                                          ⁢                  1                                                  α                  ⁢                                                                          ⁢                  2                                            ⁢                                                                    i                    O                                    ⁡                                      (                                          ω                      ⁢                                                                                          ⁢                      t                                        )                                                  ⁢                                  ⅆ                  ω                                ⁢                                                                  ⁢                t                                              =                                    1                              2                ⁢                π                                      ⁢                                          ∫                                  α                  ⁢                                                                          ⁢                  1                                                  α                  ⁢                                                                          ⁢                  2                                            ⁢                                                                    f                                          D                      ,                      T                                                        ⁡                                      (                                                                                            v                          I                                                ⁡                                                  (                                                      ω                            ⁢                                                                                                                  ⁢                            t                                                    )                                                                    -                                              V                        O                                                              )                                                  ⁢                                  ⅆ                  ω                                ⁢                                                                  ⁢                t                                                                        (        1        )            Eq. (1) is an implicit relation that defines the mean value VO of the output voltage.
The effective input waveform markedly affects proper operation of the circuit, in so far as vI appears in the integrand and affects the value of α1 and α2, which are the initial conduction angle and final conduction angle, respectively. In other words, given the same peak value, the output VO is affected in an undesirable way by the effective waveform of vI.
The dependence of VO upon the waveform applied at input is all the smaller, the closer the rectifying characteristic of the diode or of the transistor approaches the ideal situation. For this reason bipolar devices are better than FETs (Field-Effect Transistors), given their exponential transconductance.
As will be shown hereinafter, the error committed in the detection is greater if the input voltage has a higher crest factor.
In order to improve the accuracy of peak detection, described in J. Millman, A. Grabel, “Microelettronica”, McGrawHill (1994), p. 834, is a detector belonging to the class of feedback detectors. The circuit of the feedback detector is shown in FIG. 2. A high-gain stage A is used for rendering the transconductance of the rectifier element closer to the ideal situation. In correct conditions of operation, the current iO is other than zero only for a few instants around the peak of vI, compensating for the discharge current throughout the period. However, the solution is practicable only at relatively low frequencies, an additional propagation delay being present due to the gain stage.
It should be pointed out that in the circuits of FIGS. 1 and 2 the generator IDIS is sometimes replaced by a resistor, a fact that changes only to a minimal extent the equations obtained and in no way solves the problem in question.